hyperband | Tuning hyperparams fast with Hyperband | Machine Learning library

I want to code some hyperband tuning in mlr3. I started with running the subsample-rpart hyperband example from chapter 4.4 in mlr3 book - directly copied from there. I am getting an error: Error in benchmark(design, store_models = self$store_models, allow_hotstart = self$allow_hotstart, : unused argument (clone = character())

How do I fix it?

I am using Keras tuner. For the simple following code:

I am using an LSTM model in Keras. During the fitting stage, I added the validation_data paramater. When I plot my training vs validation loss, it seems there are major overfitting issues. My validation loss just won't decrease.

My full data is a sequence with shape [50,]. The first 20 records are used as training and the remaining used for the test data.

I have tried adding dropout and reducing the model complexity as much as I can and still no luck.

I am using Tensorflow's flow_from_directory to collect a large image dataset and then train on it. I want to use Keras Tuner but when I run

I have a directory of images and am taking them in like this:

I am using a Keras sequential model with LSTM layers with timeseries data to predict future values. For this I have split my data in training and validation data at a certain point in time. The timeseries data has a positive trend, thus the average values in my training data are lower then in my validation data as I am using the more recent data as validation.

The initial model predicts every time 0.5, which is a bad model. In the next epoch the model will learn by the training data and predict values on average lower than 0.5 which decreases training loss but increases validation loss. Only after a lot of epochs I will see to start a decreasing behavior for the validation loss and after even more epochs the validation loss will be lower for the first time than the first bad always prediction 0.5 model.

I am using Keras tuner with the Hyperband tuner for hyperparameter validation. This does not work for this timeseries as in its first rounds all models will show higher validation loss then the bad initial 0.5 model.

Is there a way to handle trend in timeseries in combination with Keras and splitting training and validation data? It is not possible for me to shuffle the timeseries and then split the data as I would really like to use the more recent data for validation.

I am trying to import a directory full of images into Tensorflow and then use it for Keras Tuner. The problem is Keras Tuner requires the data to be split into images and labels. I was following a guide on Tensorflow's website and here is the code I have so far:

NOTE: I am using the COCO dataset meaning each image has multiple labels. Maybe that is the problem.

I have an existing CNN model which works fine and the code is as follows.